RTX Spark amb Windows: 128 GB per a IA local, però a quin preu?

The number of partners that are so excited to bring RTX Spark to the market is just incredible. Microsoft and Nvidia is reinventing all of PC the first across the lineup for 40 years. >> Okay, Jensen, calm down. That right there was Nvidia announcing the new RTX Spark. And according to the keynote, it's going to reinvent the PC for the age of AI. >> Now, you'd think that brand new Nvidia AI machine would have everybody losing their minds, right? >> Well, instead, we get this. >> I strongly feel it should have Linux support from day one. >> Any word on Linux support? >> No, there isn't Linux support yet. And the worst thing I can imagine doing with 128 GB unified RAM is filling it up with a Windows operating system and apps. What a waste. And look, I get it. The second people hear that this thing ships with Windows out of the box, half the internet just checks out. But here's the thing, I'm not willing to say that all is lost before it even starts. In fact, I think Windows might actually be the smarter move here. >> Woah. >> To start with to start with. Now, hear me out. Let's take a look at what they actually are shipping and who it's for. >> And then you can decide if the waste crowd has a point. >> So, these days I'm always flipping between models. GPT for research, Claude for coding, Nano Banana for image

generation, Vio Kling and Runway for video. Six tabs, six bills and counting. Enter Chat LLM Teams. One dashboard houses every top LLM and route LLM picks the right one. GPT Mini for ultra fast answers, Claude Sonnet for coding, Gemini Pro for massive context. They recently added Gemini 3 and GPT 5.1 the moment they dropped. Create professional presentations with graphs, charts, and deep research detailed content. Need human-sounding copy? Humanize rewrites text to defeat AI detectors. Need visuals? Pick Frontier or open-source models. Nano Banana, Midjourney, and Flux for images, Magnific for upscaling, plus Vio Wan and Sora for video. All built-in. You also get Abacus AI deep agent to pretty much do anything. Build full-stack apps, websites, reports with just text prompts, and deploy them on the spot. They have Abacus AI desktop, which is the brand new coding editor and assistant that lets you vibe code and build production-ready apps. And the kicker? It's just $10 a month, less than one premium model. Head over to chat.abacus.ai or click the link below to level up with chat LLM teams. Now, I'm a developer and I'm a daily Mac guy who pretty much lives in the terminal. I get the reflex, but I'm going to defend Windows first here because the thing is, you need to actually examine who this machine is for. The DGX Spark and machines like it, like the Dell GB10 and other GB10s, these actually were quite a success. Wild considering it's built for a super

niche crowd, AI developers and enthusiasts. The RTX Spark is going to be a different animal. It's a much wider range of configs and it's coming to laptops and desktops. Way more people can actually buy in. Quick specs, if you don't know yet, it's up to a 20-core ARM CPU and a GPU that, on paper, reads like a 5070 RTX 5070 Ti. The RTX Spark goes up to 6,144, same as the DGX Spark, plus up to 128 gigs of unified memory that the GPU can pull from for AI. And those numbers should look pretty familiar because they're basically all the DGX Spark. Nobody on stage said it, but I will. It's the basically not a new chip. It's the same chiplet trick, a Blackwell GPU die fused to a MediaTek CPU die over NVLink. Same 20-core layout, same GPU, same core counts, but we have some flexibility there, and that comes into the different N1 versus N1X that you might have heard of that are now RTX Spark. I'll get into that momentarily. And Nvidia showed us the silicon last year. They just called it the DGX Spark. RTX Spark is just the part that's reskinned for the rest of the world. The 5070 Ti is a discrete card with its own power and its own fast VRAM. The RTX box, let's imagine that this is it right here, but Jensen showed off a MSI box I believe on stage. It's got an iGPU there sharing system memory in a thin laptop

or small factor desktop for a fraction of the watts. Based on my DGX Spark testing, I bet real world is going to land somewhere closer to a high-end AMD GPU like a Strix Halo machine than an actual 5070 Ti. I'd love to be wrong here, but I can't solve that from a slide. So, the day that these things drop, I'm going to put them head-to-head against the AMDs and check them out. And by the way, none of this is a knock. The chip is good and putting it in machines that normal people can buy is exactly the right move here. I just want to be clear that this isn't a fall out of the sky, it's a smart repackaging job. So, the question I care about isn't what the chip is, it's who Nvidia repackaged it for because that answer has changed. With the DGX Spark, the answer was pretty simple, AI developers. That's it. That was the entire audience. This time, Nvidia's going after gamers, creatives, developers, and of course the AI crowd. Basically, the whole high-performance market, not your grandma's Instagram browsing ultrabook. So, [music] first up, let's talk about gamers. Not a gaming channel here, so there's going to be other channels that cover this extensively, I'm sure. But from my perspective, who does just a little bit of I play some Doom, okay? Why would a gamer want this? One word, portable. Look at today's laptops with discreet Nvidia GPU. 5070, 5080, 5090. This Asus has a 5090 in it. It's big,

it's bulky, it's chained to a giant power brick. It's really good for gaming. Really good. But an RTX Spark machine could be thin and light instead. This is a MacBook, by the way. Battery's going to be the big question here. My DGX Spark pulls up to about 140 watts when I'm really pushing it. Some leaks mention that the RTX Spark envelope is way lower than that, 45 to 80 watts depending on the config. If you take the lower end of that, 45 watts, that's good news. That is a chip that could last a real work day on battery. More thin and light than a gaming brick, but push it towards 80 watts or higher and I get a little nervous because I felt how toasty the DGX Spark gets when you're pushing it to the max. And that thing has a big fan in it, well, relatively big. I mean, it's a small box, right? It's got a lot more room to dissipate the heat than laptop has. Will the laptops throttle if you push it? I don't know, it's an open question, of course. I'll test both ends when I get one or a couple. We'll see. Now, gamers, where do most gamers live? Windows. That's the whole ballgame here. But those aren't the only people that live in Windows. Creatives, the people that live in Adobe world, Photoshop, Premiere, After Effects. These are tools that are brutally memory and power hungry. When I'm editing videos, I regularly hit 90

GB used on my MacBook. And creatives need like 32, 64, 128 gigs of RAM. They could grab a MacBook Pro, that gets the job done. Plays nice with creative apps. Power usage is pretty good on those and they last a long time. So, creatives is probably their weakest case for the RTX Spark. But if their job requires a PC, it's Windows. These suites don't run on Linux. To this day, after many, many years of begging Adobe to come to the Linux world, they're just they're just not doing it. And here's the part that bugs me the most. It's the perfect Windows versus Linux illustration because Adobe re-engineered the core part of Photoshop and Premiere, not only for the RTX Spark that's coming up, that runs on Windows on ARM. They did this for Qualcomm's XLE chips a couple years ago. They already did the hard work, but there's still no Premiere Pro for Linux. My people, developers. [music] Here's a config for just about all of us. There it comes in everything from 16 GB up to 128 GB. So, web devs, mobile devs, you name it, there's a fit. Here's the wrinkle though for devs. We're leaning more and more on Linux these days. Sure, the .NET and Visual Studio world that I actually grew up on runs great on Windows for ARM now, and that's been solid for a couple years. Now, Nvidia can ship the hardware and Microsoft can nail the OS support, although nail is

probably um too exact of a word. Staple gun? Maybe multiple attempts? I'm being too harsh. Windows on ARM is much better than it was. But whether the tools you actually depend on show up on your platform is a vendor-by-vendor call. Nvidia can't just buy its way through, but they can try. But in the end, great silicon doesn't help if the software you live in won't run on the OS that you want. So, I think, this is just my theory, the devs are going to be split. Some are going to be on Linux boxes, some are on MacBooks, some on x86 Windows machines, cuz their employers says, "Here you go." And don't forget Qualcomm's already in the Windows on ARM space. Snapdragon X Elite, Snapdragon X2 Elite, with extreme, seriously powerful, ridiculous battery, thin laptops. If I weren't living on a MacBook every day, I'd probably be on an X2 Elite right now. But the RTX machine's trump card is CUDA. If your work lives on the CUDA stack, you're going to be getting one of those. And that leads straight into the last group of people. AI developers [music] and enthusiasts. The DGX Spark proved people want these things to be portable. You can put this in a portable case and bring it with you. Just get one of these. Don't forget the power brick, the monitor, the cables, the mouse, the keyboard, all that stuff. These are CUDA-based AI machines, and RTX Spark just lets Nvidia grab way more of that market. Let's talk

about money for a moment. This is where I get off the hype train here. The DGX Spark was initially thought to be $3,000, then at launch it became $4,000, and then it became 46 to 4700. The Dell version is even more than that. That's not enthusiast money, that's justified on your tax form money. But, I think there's a couple of differences here that are key. Big models want big memory and big storage. So, the serious crowd for AI people, those are going to be maxing out. 128, 4 terabytes, whatever they'll come with. I don't know what they'll come with. I just know the memory, and the bill is going to climb fast. But, on the flip side, with the thinner, lighter configurations, like a base config, maybe that's going to be 16 GB memory, maybe a terabyte or 512, don't you dare, but maybe 512. I don't know, I hope not. Terabyte, please. Base config, you could also lean on external storage because if it has USB 4 on there, you can save a few hundred bucks and get yourself external storage for your models. Totally reasonable. Here's another thing that might affect the pricing. Remember when I said up to 20 cores? The hero chip is basically going to be pretty much the same one that's in the DGX Spark. It's a full 20-core CPU, but there's going to be a cut-down version. Maybe the GPU will also be cut down to a smaller amount of cores. It's

basically binned silicon. Chips are made on a silicon circular wafer, and some around the edges maybe don't make it. They get cut off. So, some chips come off the line with a dead core or maybe bad GPU blocks. They're switched off and they're sold cheaper. That's not a shady practice, by the way. Everybody bins, Apple, AMD included. So, that's where the N1 and the N1X come from. N1X is the full spec, N1 is the lower spec, but we don't know exactly for sure what is what at this point yet. The N1 might go down to as low as like eight or 10 cores for the CPU, which is going to bring the price down. So, an RTX Spark could mean a 128 gig 20-core monster or a 16-gig chip running a third of the cores. Same name, very different machines, which is why I'll keep mentioning, "Well, which config are you talking about when you're talking about the pricing here?" I've already seen a bunch of people commenting and tweeting that this will cost as much as the DJ X fire core more. Well, hold on. Remember what's actually inside of these GB10 boxes. There's a dual ConnectX 7 interface. That's a pro-grade data center networking. And that part alone is something like 1,500 bucks. The RTX Sparks are going to be consumer machines. So, they'll almost certainly ship with something like a plain 10-gig Ethernet instead. Now, let's get back to the Windows question. Microsoft, of course, will probably push

this Co-pilot first like they did with Snapdragon, but I got to admit, I'm going to I'm going to come out. I'm going to come out and say this. After doing all these AI tests and working with all these Snapdragon laptops and all these Co-pilot PCs, I've not once, not once tried Co-pilot. You can do a lot of AI on a machine and use its hardware and use its software stack without touching Co-pilot. But, Microsoft's job is to try and push that on people, and they won't stop. They won't stop. Anyway, the real AI work still leans heavily on Linux. And I think that Nvidia, with its giant bags of cash, can get into Linux way faster than Qualcomm ever managed. So, there's hope. So, basically, why do I think that Windows first is actually a good thing? Well, every single group that I mentioned lives in Windows first. But, Windows for ARM? It's also an uphill battle. It's just like what we saw with Qualcomm a couple years ago in X Elite. App makers only port to ARM if there's an audience waiting. It's like a catch-22 kind of thing. But, we just met the audience. And I think that this audience, the Nvidia set, is actually going to be larger than the Qualcomm set because of CUDA. So, in my opinion, Windows first is the fastest path to a real arm ecosystem. Native Adobe, native games, working drivers, and that ecosystem is exactly what Linux on arm

inherits later. The people that are screaming, "We want Linux now." they're just mad about the move that's actually quietly working in their favor. It just takes a little bit more patience. And nobody mentions this. Windows is actually a difficult case. So, they led with the hard problem. Linux on arm is basically solved. The default in servers and phones for years. Getting Windows with its emulation and 40-year app catalog to run well on arm is the brutal part here. Qualcomm and Microsoft have been grinding on Windows for arm for years now. Prism, the emulation layer, the native boards with the app developers, the X Elite laptops that I've been testing a lot on this channel. So, Nvidia isn't betting on an unproven OS. They're just dropping CUDA, the one thing that nobody else has, onto a platform that's already been beaten into shape. Mostly. That's actually a good gamble and good timing. And again, to the people that want Linux right now, you already have a machine. It's called the DGX Spark. It runs Linux right now. It's not going anywhere and in fact it's going to inherit the new chips like Vera Rubin, Feynman with whatever the CPU is that's matched up with that. That was mentioned in the keynote, too. So, is it all a waste? Based on what I'm looking at? No, I don't think so. And I want to be clear that it's a judgment on the strategy, not on the hardware because none of us

have benched the hardware yet. Sure, there's some Geekbench numbers of the N1X have been around for over a year now at this point. But those are early numbers and they're not the real numbers. I've been excited about Windows on arm and you can go back in my videos and look through the entire couple of years that we've had Snapdragon machines. And now finally there's going to be some real competition and some company for X Elites. That part is good for everybody. Is it a smartphone moment? I don't know. That's a heck of a thing to claim on stage before anyone lived with one, but Jensen goes big and he makes big bold claims. And you know what? For the most part, he was kind of right about the Spark. I think RTX Spark is going to be big. Is it going to be a new era of PC like everybody claims? Probably not, but it's definitely not a waste. Tell me down below what you think of my Windows first analysis. If you agree, disagree, I'm sure I'm going to have a lot of disagreements. And if you're curious about how the DGX Spark performs, here's a couple of videos for you to watch right here. Thanks [music] for watching and I'll see you next time.